id stringlengths 12 32 | archetype stringclasses 30
values | split stringclasses 1
value | prompt stringlengths 4.71k 5.31k | gold stringlengths 68 186 | meta stringclasses 30
values |
|---|---|---|---|---|---|
train-training_time-064 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.7111111111111111}, "work": {"total_tokens": 81920000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-accum_steps_lr-019 | accum_steps_lr | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_STEPS_AND_LR", "fix": {"current_learning_rate": 0.00011, "grad_accum_steps": 1, "planned_total_steps": 750}, "work": {"audit_step": 375, "steps_per_epoch": 250}} | {"archetype": "accum_steps_lr", "computations": ["R3", "R14", "R15"], "depth": 5, "interaction": "accum x schedule", "tier": "interacting"} |
train-ckpt_batch_accum-007 | ckpt_batch_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINTED_BATCH_AND_ACCUM", "fix": {"grad_accum_steps": 8, "per_device_batch_size": 37}, "work": {"activation_per_sample_gb": 2.0, "max_per_device_batch": 37}} | {"archetype": "ckpt_batch_accum", "computations": ["R20", "R1", "R3"], "depth": 4, "interaction": "checkpointing x accum", "tier": "interacting"} |
train-shard_size-006 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 5000, "shard_size_gb": 60.0}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-two_sampler_accum-003 | two_sampler_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["DUPLICATE_DATA_NO_DISTRIBUTED_SAMPLER", "EFFECTIVE_BATCH_MISMATCH"], "fix": {"grad_accum_steps": 16, "sampler": "DistributedSampler"}} | {"archetype": "two_sampler_accum", "computations": ["R7", "R3"], "depth": 3, "interaction": "sampler+accum", "tier": "two_fault"} |
train-shard_size-028 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 5000, "shard_size_gb": 40.0}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-two_oom_clipping-018 | two_oom_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["OOM_BATCH_TOO_LARGE", "NO_GRADIENT_CLIPPING"], "fix": {"max_grad_norm": 1.0, "per_device_batch_size": 35}} | {"archetype": "two_oom_clipping", "computations": ["R1", "R12"], "depth": 3, "interaction": "oom+clipping", "tier": "two_fault"} |
train-packing_time-002 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 0.5}, "work": {"total_padded_tokens": 163840000, "total_real_tokens": 144000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-vocab-000 | vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "EMBEDDING_VOCAB_MISMATCH", "fix": {"embedding_vocab_size": 64032}} | {"archetype": "vocab", "computations": ["R11"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-batch_from_components-067 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 10}, "work": {"fixed_memory_gb": 64.0, "remaining_memory_gb": 16.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-batch_from_components-013 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 32}, "work": {"fixed_memory_gb": 48.0, "remaining_memory_gb": 32.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_v100_dataloader-031 | two_v100_dataloader | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["PRECISION_UNSUPPORTED_ON_GPU", "DATALOADER_FORK_WITH_CUDA"], "fix": {"dataloader_multiprocessing_context": "spawn", "dtype": "fp16", "use_grad_scaler": true}} | {"archetype": "two_v100_dataloader", "computations": ["R5", "R9"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-accum_and_steps-008 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 4, "planned_total_steps": 1200}, "work": {"steps_per_epoch": 400}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-lr_at_step-028 | lr_at_step | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "LR_VALUE_AT_STEP", "fix": {"current_learning_rate": 0.000155}, "work": {"step_fraction": 0.25}} | {"archetype": "lr_at_step", "computations": ["R15"], "depth": 2, "interaction": null, "tier": "computed"} |
train-shard_size-055 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 25000, "shard_size_gb": 25.0}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-accum_and_steps-032 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 2, "planned_total_steps": 600}, "work": {"steps_per_epoch": 200}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-shard_size-003 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 5000, "shard_size_gb": 60.0}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-accum_steps_lr-003 | accum_steps_lr | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_STEPS_AND_LR", "fix": {"current_learning_rate": 0.000465, "grad_accum_steps": 8, "planned_total_steps": 750}, "work": {"audit_step": 187, "steps_per_epoch": 250}} | {"archetype": "accum_steps_lr", "computations": ["R3", "R14", "R15"], "depth": 5, "interaction": "accum x schedule", "tier": "interacting"} |
train-training_time-017 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.7111111111111111}, "work": {"total_tokens": 204800000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-shard_size-029 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 12500, "shard_size_gb": 40.0}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-parallel_batch-000 | parallel_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PARALLEL_SHARDED_BATCH", "fix": {"per_device_batch_size": 64}, "work": {"fixed_memory_gb": 64.0, "params_per_rank_billions": 4.0}} | {"archetype": "parallel_batch", "computations": ["R22", "R13", "R1"], "depth": 4, "interaction": "parallelism x memory", "tier": "interacting"} |
train-training_time-010 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 1.0666666666666667}, "work": {"total_tokens": 307200000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-batch_from_components-036 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 5}, "work": {"fixed_memory_gb": 32.0, "remaining_memory_gb": 8.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-checkpoint_retention-003 | checkpoint_retention | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINT_RETENTION_LIMIT", "fix": {"max_checkpoints_to_keep": 6}, "work": {"checkpoint_size_gb": 112.0}} | {"archetype": "checkpoint_retention", "computations": ["R17"], "depth": 2, "interaction": null, "tier": "computed"} |
train-training_time-031 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.7111111111111111}, "work": {"total_tokens": 245760000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_oom_clipping-014 | two_oom_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["OOM_BATCH_TOO_LARGE", "NO_GRADIENT_CLIPPING"], "fix": {"max_grad_norm": 1.0, "per_device_batch_size": 9}} | {"archetype": "two_oom_clipping", "computations": ["R1", "R12"], "depth": 3, "interaction": "oom+clipping", "tier": "two_fault"} |
train-optimizer_batch-032 | optimizer_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "OPTIMIZER_MEMORY_BATCH", "fix": {"per_device_batch_size": 10}, "work": {"bytes_per_param": 12, "fixed_memory_gb": 24.0}} | {"archetype": "optimizer_batch", "computations": ["R21", "R1"], "depth": 3, "interaction": "optimizer x memory", "tier": "interacting"} |
train-lr_at_step-001 | lr_at_step | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "LR_VALUE_AT_STEP", "fix": {"current_learning_rate": 0.00011}, "work": {"step_fraction": 0.5}} | {"archetype": "lr_at_step", "computations": ["R15"], "depth": 2, "interaction": null, "tier": "computed"} |
train-optimizer_batch-025 | optimizer_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "OPTIMIZER_MEMORY_BATCH", "fix": {"per_device_batch_size": 69}, "work": {"bytes_per_param": 8, "fixed_memory_gb": 24.0}} | {"archetype": "optimizer_batch", "computations": ["R21", "R1"], "depth": 3, "interaction": "optimizer x memory", "tier": "interacting"} |
train-checkpointing-001 | checkpointing | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACTIVATION_MEMORY_NO_CHECKPOINTING", "fix": {"gradient_checkpointing": true}} | {"archetype": "checkpointing", "computations": ["R2"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-packing_time-028 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 1.0}, "work": {"total_padded_tokens": 163840000, "total_real_tokens": 144000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-two_sampler_accum-007 | two_sampler_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["DUPLICATE_DATA_NO_DISTRIBUTED_SAMPLER", "EFFECTIVE_BATCH_MISMATCH"], "fix": {"grad_accum_steps": 4, "sampler": "DistributedSampler"}} | {"archetype": "two_sampler_accum", "computations": ["R7", "R3"], "depth": 3, "interaction": "sampler+accum", "tier": "two_fault"} |
train-accum_and_steps-053 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 4, "planned_total_steps": 300}, "work": {"steps_per_epoch": 100}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-training_time-063 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.11851851851851852}, "work": {"total_tokens": 40960000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-accum_steps_lr-010 | accum_steps_lr | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_STEPS_AND_LR", "fix": {"current_learning_rate": 0.000465, "grad_accum_steps": 16, "planned_total_steps": 200}, "work": {"audit_step": 50, "steps_per_epoch": 100}} | {"archetype": "accum_steps_lr", "computations": ["R3", "R14", "R15"], "depth": 5, "interaction": "accum x schedule", "tier": "interacting"} |
train-parallel_batch-060 | parallel_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PARALLEL_SHARDED_BATCH", "fix": {"per_device_batch_size": 25}, "work": {"fixed_memory_gb": 128.0, "params_per_rank_billions": 8.0}} | {"archetype": "parallel_batch", "computations": ["R22", "R13", "R1"], "depth": 4, "interaction": "parallelism x memory", "tier": "interacting"} |
train-parallel_batch-023 | parallel_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PARALLEL_SHARDED_BATCH", "fix": {"per_device_batch_size": 24}, "work": {"fixed_memory_gb": 32.0, "params_per_rank_billions": 2.0}} | {"archetype": "parallel_batch", "computations": ["R22", "R13", "R1"], "depth": 4, "interaction": "parallelism x memory", "tier": "interacting"} |
train-batch_from_components-027 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 32}, "work": {"fixed_memory_gb": 48.0, "remaining_memory_gb": 32.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-optimizer_batch-049 | optimizer_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "OPTIMIZER_MEMORY_BATCH", "fix": {"per_device_batch_size": 12}, "work": {"bytes_per_param": 12, "fixed_memory_gb": 84.0}} | {"archetype": "optimizer_batch", "computations": ["R21", "R1"], "depth": 3, "interaction": "optimizer x memory", "tier": "interacting"} |
train-training_time-006 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.28444444444444444}, "work": {"total_tokens": 122880000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-batch_from_components-059 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 24}, "work": {"fixed_memory_gb": 32.0, "remaining_memory_gb": 48.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-batch_from_components-008 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 32}, "work": {"fixed_memory_gb": 32.0, "remaining_memory_gb": 48.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-optimizer_batch-039 | optimizer_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "OPTIMIZER_MEMORY_BATCH", "fix": {"per_device_batch_size": 16}, "work": {"bytes_per_param": 8, "fixed_memory_gb": 56.0}} | {"archetype": "optimizer_batch", "computations": ["R21", "R1"], "depth": 3, "interaction": "optimizer x memory", "tier": "interacting"} |
train-shard_size-063 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 5000, "shard_size_gb": 12.5}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-lr_at_step-038 | lr_at_step | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "LR_VALUE_AT_STEP", "fix": {"current_learning_rate": 0.000165}, "work": {"step_fraction": 0.5}} | {"archetype": "lr_at_step", "computations": ["R15"], "depth": 2, "interaction": null, "tier": "computed"} |
train-trap_nan_clipping-006 | trap_nan_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "NO_GRADIENT_CLIPPING", "fix": {"max_grad_norm": 1.0}} | {"archetype": "trap_nan_clipping", "computations": ["R4", "R12"], "depth": 2, "interaction": "lr", "tier": "trap"} |
train-accum_and_steps-033 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 4, "planned_total_steps": 600}, "work": {"steps_per_epoch": 200}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_v100_dataloader-009 | two_v100_dataloader | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["PRECISION_UNSUPPORTED_ON_GPU", "DATALOADER_FORK_WITH_CUDA"], "fix": {"dataloader_multiprocessing_context": "spawn", "dtype": "fp16", "use_grad_scaler": true}} | {"archetype": "two_v100_dataloader", "computations": ["R5", "R9"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-two_lora_vocab-051 | two_lora_vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["ADAPTERS_FROZEN_ZERO_TRAINABLE", "EMBEDDING_VOCAB_MISMATCH"], "fix": {"adapter_requires_grad": true, "embedding_vocab_size": 128256}} | {"archetype": "two_lora_vocab", "computations": ["R8", "R11"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-accum_and_steps-065 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 8, "planned_total_steps": 400}, "work": {"steps_per_epoch": 100}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-batch_from_components-063 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 37}, "work": {"fixed_memory_gb": 24.0, "remaining_memory_gb": 56.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-training_time-033 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.9481481481481482}, "work": {"total_tokens": 409600000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_v100_dataloader-017 | two_v100_dataloader | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["PRECISION_UNSUPPORTED_ON_GPU", "DATALOADER_FORK_WITH_CUDA"], "fix": {"dataloader_multiprocessing_context": "spawn", "dtype": "fp16", "use_grad_scaler": true}} | {"archetype": "two_v100_dataloader", "computations": ["R5", "R9"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-trap_oom_checkpointing-006 | trap_oom_checkpointing | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACTIVATION_MEMORY_NO_CHECKPOINTING", "fix": {"gradient_checkpointing": true}} | {"archetype": "trap_oom_checkpointing", "computations": ["R1", "R2"], "depth": 2, "interaction": "oom_batch", "tier": "trap"} |
train-lr_at_step-033 | lr_at_step | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "LR_VALUE_AT_STEP", "fix": {"current_learning_rate": 0.000195}, "work": {"step_fraction": 0.75}} | {"archetype": "lr_at_step", "computations": ["R15"], "depth": 2, "interaction": null, "tier": "computed"} |
train-ckpt_batch_accum-019 | ckpt_batch_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINTED_BATCH_AND_ACCUM", "fix": {"grad_accum_steps": 2, "per_device_batch_size": 17}, "work": {"activation_per_sample_gb": 1.5, "max_per_device_batch": 17}} | {"archetype": "ckpt_batch_accum", "computations": ["R20", "R1", "R3"], "depth": 4, "interaction": "checkpointing x accum", "tier": "interacting"} |
train-ckpt_batch_accum-002 | ckpt_batch_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINTED_BATCH_AND_ACCUM", "fix": {"grad_accum_steps": 4, "per_device_batch_size": 15}, "work": {"activation_per_sample_gb": 2.0, "max_per_device_batch": 15}} | {"archetype": "ckpt_batch_accum", "computations": ["R20", "R1", "R3"], "depth": 4, "interaction": "checkpointing x accum", "tier": "interacting"} |
train-zero_batch-055 | zero_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ZERO_SHARDED_BATCH_SIZE", "fix": {"per_device_batch_size": 80}, "work": {"bytes_per_param_per_rank": 12.0, "fixed_memory_gb": 48.0, "remaining_memory_gb": 80.0}} | {"archetype": "zero_batch", "computations": ["R19", "R13", "R1"], "depth": 4, "interaction": "zero x memory", "tier": "interacting"} |
train-accum_and_steps-068 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 2, "planned_total_steps": 400}, "work": {"steps_per_epoch": 200}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-batch_from_components-025 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 6}, "work": {"fixed_memory_gb": 24.0, "remaining_memory_gb": 16.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-training_time-037 | training_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "TRAINING_TIME_ESTIMATE", "fix": {"planned_time_hours": 0.7111111111111111}, "work": {"total_tokens": 204800000}} | {"archetype": "training_time", "computations": ["R16"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_lora_vocab-017 | two_lora_vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["ADAPTERS_FROZEN_ZERO_TRAINABLE", "EMBEDDING_VOCAB_MISMATCH"], "fix": {"adapter_requires_grad": true, "embedding_vocab_size": 128256}} | {"archetype": "two_lora_vocab", "computations": ["R8", "R11"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-clipping-005 | clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "NO_GRADIENT_CLIPPING", "fix": {"max_grad_norm": 1.0}} | {"archetype": "clipping", "computations": ["R12"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-batch_from_components-002 | batch_from_components | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "BATCH_FROM_MEMORY_COMPONENTS", "fix": {"per_device_batch_size": 16}, "work": {"fixed_memory_gb": 48.0, "remaining_memory_gb": 32.0}} | {"archetype": "batch_from_components", "computations": ["R13", "R1"], "depth": 3, "interaction": null, "tier": "computed"} |
train-optimizer_batch-018 | optimizer_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "OPTIMIZER_MEMORY_BATCH", "fix": {"per_device_batch_size": 6}, "work": {"bytes_per_param": 12, "fixed_memory_gb": 84.0}} | {"archetype": "optimizer_batch", "computations": ["R21", "R1"], "depth": 3, "interaction": "optimizer x memory", "tier": "interacting"} |
train-two_oom_clipping-058 | two_oom_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["OOM_BATCH_TOO_LARGE", "NO_GRADIENT_CLIPPING"], "fix": {"max_grad_norm": 1.0, "per_device_batch_size": 26}} | {"archetype": "two_oom_clipping", "computations": ["R1", "R12"], "depth": 3, "interaction": "oom+clipping", "tier": "two_fault"} |
train-packing_time-040 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 0.5}, "work": {"total_padded_tokens": 491520000, "total_real_tokens": 72000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-checkpoint_retention-008 | checkpoint_retention | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINT_RETENTION_LIMIT", "fix": {"max_checkpoints_to_keep": 31}, "work": {"checkpoint_size_gb": 24.0}} | {"archetype": "checkpoint_retention", "computations": ["R17"], "depth": 2, "interaction": null, "tier": "computed"} |
train-accum_and_steps-037 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 4, "planned_total_steps": 400}, "work": {"steps_per_epoch": 100}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_lora_vocab-055 | two_lora_vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["ADAPTERS_FROZEN_ZERO_TRAINABLE", "EMBEDDING_VOCAB_MISMATCH"], "fix": {"adapter_requires_grad": true, "embedding_vocab_size": 32064}} | {"archetype": "two_lora_vocab", "computations": ["R8", "R11"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-two_oom_clipping-025 | two_oom_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["OOM_BATCH_TOO_LARGE", "NO_GRADIENT_CLIPPING"], "fix": {"max_grad_norm": 1.0, "per_device_batch_size": 13}} | {"archetype": "two_oom_clipping", "computations": ["R1", "R12"], "depth": 3, "interaction": "oom+clipping", "tier": "two_fault"} |
train-two_sampler_accum-059 | two_sampler_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["DUPLICATE_DATA_NO_DISTRIBUTED_SAMPLER", "EFFECTIVE_BATCH_MISMATCH"], "fix": {"grad_accum_steps": 8, "sampler": "DistributedSampler"}} | {"archetype": "two_sampler_accum", "computations": ["R7", "R3"], "depth": 3, "interaction": "sampler+accum", "tier": "two_fault"} |
train-packing_time-005 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 0.5}, "work": {"total_padded_tokens": 163840000, "total_real_tokens": 144000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-trap_oom_checkpointing-009 | trap_oom_checkpointing | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACTIVATION_MEMORY_NO_CHECKPOINTING", "fix": {"gradient_checkpointing": true}} | {"archetype": "trap_oom_checkpointing", "computations": ["R1", "R2"], "depth": 2, "interaction": "oom_batch", "tier": "trap"} |
train-two_oom_clipping-055 | two_oom_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["OOM_BATCH_TOO_LARGE", "NO_GRADIENT_CLIPPING"], "fix": {"max_grad_norm": 1.0, "per_device_batch_size": 26}} | {"archetype": "two_oom_clipping", "computations": ["R1", "R12"], "depth": 3, "interaction": "oom+clipping", "tier": "two_fault"} |
train-lr_at_step-035 | lr_at_step | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "LR_VALUE_AT_STEP", "fix": {"current_learning_rate": 0.000155}, "work": {"step_fraction": 0.25}} | {"archetype": "lr_at_step", "computations": ["R15"], "depth": 2, "interaction": null, "tier": "computed"} |
train-vocab-004 | vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "EMBEDDING_VOCAB_MISMATCH", "fix": {"embedding_vocab_size": 32064}} | {"archetype": "vocab", "computations": ["R11"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-trap_nan_clipping-009 | trap_nan_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "NO_GRADIENT_CLIPPING", "fix": {"max_grad_norm": 1.0}} | {"archetype": "trap_nan_clipping", "computations": ["R4", "R12"], "depth": 2, "interaction": "lr", "tier": "trap"} |
train-clipping-004 | clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "NO_GRADIENT_CLIPPING", "fix": {"max_grad_norm": 1.0}} | {"archetype": "clipping", "computations": ["R12"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-accum_and_steps-048 | accum_and_steps | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACCUM_AND_TOTAL_STEPS", "fix": {"grad_accum_steps": 2, "planned_total_steps": 200}, "work": {"steps_per_epoch": 100}} | {"archetype": "accum_and_steps", "computations": ["R3", "R14"], "depth": 3, "interaction": null, "tier": "computed"} |
train-two_v100_dataloader-026 | two_v100_dataloader | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["PRECISION_UNSUPPORTED_ON_GPU", "DATALOADER_FORK_WITH_CUDA"], "fix": {"dataloader_multiprocessing_context": "spawn", "dtype": "fp16", "use_grad_scaler": true}} | {"archetype": "two_v100_dataloader", "computations": ["R5", "R9"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-trap_oom_checkpointing-002 | trap_oom_checkpointing | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "ACTIVATION_MEMORY_NO_CHECKPOINTING", "fix": {"gradient_checkpointing": true}} | {"archetype": "trap_oom_checkpointing", "computations": ["R1", "R2"], "depth": 2, "interaction": "oom_batch", "tier": "trap"} |
train-vocab-006 | vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "EMBEDDING_VOCAB_MISMATCH", "fix": {"embedding_vocab_size": 50304}} | {"archetype": "vocab", "computations": ["R11"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-clipping-007 | clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "NO_GRADIENT_CLIPPING", "fix": {"max_grad_norm": 1.0}} | {"archetype": "clipping", "computations": ["R12"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-packing_time-031 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 1.0}, "work": {"total_padded_tokens": 245760000, "total_real_tokens": 216000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-device_mismatch-005 | device_mismatch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "DEVICE_MISMATCH_LOSS", "fix": {"loss_device": "cuda:0"}} | {"archetype": "device_mismatch", "computations": ["R10"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-two_sampler_accum-028 | two_sampler_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["DUPLICATE_DATA_NO_DISTRIBUTED_SAMPLER", "EFFECTIVE_BATCH_MISMATCH"], "fix": {"grad_accum_steps": 8, "sampler": "DistributedSampler"}} | {"archetype": "two_sampler_accum", "computations": ["R7", "R3"], "depth": 3, "interaction": "sampler+accum", "tier": "two_fault"} |
train-two_oom_clipping-056 | two_oom_clipping | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["OOM_BATCH_TOO_LARGE", "NO_GRADIENT_CLIPPING"], "fix": {"max_grad_norm": 1.0, "per_device_batch_size": 12}} | {"archetype": "two_oom_clipping", "computations": ["R1", "R12"], "depth": 3, "interaction": "oom+clipping", "tier": "two_fault"} |
train-packing_time-003 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 0.5}, "work": {"total_padded_tokens": 163840000, "total_real_tokens": 144000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-two_sampler_accum-006 | two_sampler_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["DUPLICATE_DATA_NO_DISTRIBUTED_SAMPLER", "EFFECTIVE_BATCH_MISMATCH"], "fix": {"grad_accum_steps": 8, "sampler": "DistributedSampler"}} | {"archetype": "two_sampler_accum", "computations": ["R7", "R3"], "depth": 3, "interaction": "sampler+accum", "tier": "two_fault"} |
train-lr_at_step-000 | lr_at_step | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "LR_VALUE_AT_STEP", "fix": {"current_learning_rate": 0.000195}, "work": {"step_fraction": 0.75}} | {"archetype": "lr_at_step", "computations": ["R15"], "depth": 2, "interaction": null, "tier": "computed"} |
train-bf16_v100-005 | bf16_v100 | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PRECISION_UNSUPPORTED_ON_GPU", "fix": {"dtype": "fp16", "use_grad_scaler": true}} | {"archetype": "bf16_v100", "computations": ["R5", "R6"], "depth": 1, "interaction": null, "tier": "diagnosis"} |
train-optimizer_batch-028 | optimizer_batch | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "OPTIMIZER_MEMORY_BATCH", "fix": {"per_device_batch_size": 29}, "work": {"bytes_per_param": 12, "fixed_memory_gb": 84.0}} | {"archetype": "optimizer_batch", "computations": ["R21", "R1"], "depth": 3, "interaction": "optimizer x memory", "tier": "interacting"} |
train-two_sampler_accum-002 | two_sampler_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["DUPLICATE_DATA_NO_DISTRIBUTED_SAMPLER", "EFFECTIVE_BATCH_MISMATCH"], "fix": {"grad_accum_steps": 8, "sampler": "DistributedSampler"}} | {"archetype": "two_sampler_accum", "computations": ["R7", "R3"], "depth": 3, "interaction": "sampler+accum", "tier": "two_fault"} |
train-packing_time-014 | packing_time | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "PACKED_TRAINING_TIME", "fix": {"planned_time_hours": 0.5}, "work": {"total_padded_tokens": 245760000, "total_real_tokens": 72000000}} | {"archetype": "packing_time", "computations": ["R23", "R16"], "depth": 4, "interaction": "packing x time", "tier": "interacting"} |
train-shard_size-048 | shard_size | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "SHARD_SIZE_PER_RANK", "fix": {"examples_per_rank": 10000, "shard_size_gb": 60.0}} | {"archetype": "shard_size", "computations": ["R18"], "depth": 2, "interaction": null, "tier": "computed"} |
train-ckpt_batch_accum-030 | ckpt_batch_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINTED_BATCH_AND_ACCUM", "fix": {"grad_accum_steps": 8, "per_device_batch_size": 44}, "work": {"activation_per_sample_gb": 1.5, "max_per_device_batch": 44}} | {"archetype": "ckpt_batch_accum", "computations": ["R20", "R1", "R3"], "depth": 4, "interaction": "checkpointing x accum", "tier": "interacting"} |
train-two_lora_vocab-050 | two_lora_vocab | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": ["ADAPTERS_FROZEN_ZERO_TRAINABLE", "EMBEDDING_VOCAB_MISMATCH"], "fix": {"adapter_requires_grad": true, "embedding_vocab_size": 128256}} | {"archetype": "two_lora_vocab", "computations": ["R8", "R11"], "depth": 2, "interaction": "mixed", "tier": "two_fault"} |
train-ckpt_batch_accum-021 | ckpt_batch_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINTED_BATCH_AND_ACCUM", "fix": {"grad_accum_steps": 2, "per_device_batch_size": 49}, "work": {"activation_per_sample_gb": 1.5, "max_per_device_batch": 49}} | {"archetype": "ckpt_batch_accum", "computations": ["R20", "R1", "R3"], "depth": 4, "interaction": "checkpointing x accum", "tier": "interacting"} |
train-ckpt_batch_accum-005 | ckpt_batch_accum | train | You are debugging a failed or misbehaving LLM training run.
Pick exactly ONE diagnosis code and give the minimal fix: only the config fields that must change, with corrected values.
Diagnosis codes:
- OOM_BATCH_TOO_LARGE
- ACTIVATION_MEMORY_NO_CHECKPOINTING
- EFFECTIVE_BATCH_MISMATCH
- LR_TOO_HIGH_NO_WARMUP
- PRECISIO... | {"diagnosis": "CHECKPOINTED_BATCH_AND_ACCUM", "fix": {"grad_accum_steps": 4, "per_device_batch_size": 22}, "work": {"activation_per_sample_gb": 1.5, "max_per_device_batch": 22}} | {"archetype": "ckpt_batch_accum", "computations": ["R20", "R1", "R3"], "depth": 4, "interaction": "checkpointing x accum", "tier": "interacting"} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.